OpenAI plans to bring hidden watermark for ChatGPT and Codex

OpenAI is moving towards a system that could allow detectors to identify text produced by its AI models without placing any visible label on the page. The measure will initially target eligible users in the European Union, where new rules require AI-generated material to carry identifiable signals.OpenAI is preparing to embed an invisible signature in text generated by ChatGPT and Codex for users in the European Union, giving researchers and other authorised organisations a way to test whether a passage was produced by one of its systems.